TAIPEI, Taiwan, Oct. 13, 2025 — TrendForce’s latest findings indicate that the surging demand for AI servers is prompting major global cloud service providers (CSPs), including Google, AWS, Meta, Microsoft, Oracle, Tencent, Alibaba, and Baidu, to significantly increase their investments in NVIDIA’s rack-scale GPU solutions, data center expansion, and proprietary AI ASIC development. The combined capital expenditure (CapEx) from these eight leading CSPs is projected to exceed US$420 billion in 2025, which is roughly equivalent to their total spending in 2023 and 2024, representing a 61% year-over-year increase.
TrendForce forecasts that as rack-scale solutions such as GB/VR systems continue to expand in 2026, the total CapEx of the eight CSPs will reach an unprecedented high, surpassing US$520 billion with a 24% year-over-year growth. Furthermore, investment priorities are shifting from assets that generate direct revenue to more immediate infrastructure like servers and GPUs. This signals a strategic focus on enhancing long-term competitiveness and market share rather than immediate financial gains.
In 2025, NVIDIA’s GB200/GB300 rack systems are poised to become the primary deployment choice for CSPs, driven by stronger-than-anticipated demand growth. Beyond the top four North American CSPs and Oracle, emerging clients like Tesla/xAI, CoreWeave, and Nebius are escalating their purchases for AI cloud leasing and generative AI workloads. By the latter half of 2026, CSPs are expected to transition from GB300 racks to the new NVIDIA Rubin VR200 rack platform.
Production of Custom AI Chips Continues to Expand
North America’s four leading CSPs are intensifying their investments in AI ASICs to improve autonomy and manage costs associated with large-scale AI and LLM tasks. Google, in collaboration with Broadcom, is developing the TPU v7p (Ironwood), an optimized training platform scheduled for expansion in 2026, succeeding the TPU v6e (Trilium). TrendForce predicts Google’s TPU shipments will remain the highest among CSPs, with an annual growth exceeding 40% in 2026.
AWS is prioritizing its Trainium v2 chips, with a liquid-cooled rack version expected by late 2025. The Trainium v3, co-developed with Alchip and Marvell, is slated for mass production in early 2026. AWS’s ASIC shipments are anticipated to more than double in 2025, marking the fastest growth among major CSPs, with an additional approximately 20% increase in 2026.
Meta is strengthening its partnership with Broadcom, aiming for MTIA v2 mass production in Q4 2025 to boost inference efficiency and lower latency. In 2025, shipments will primarily support Meta’s internal AI platforms and recommendation systems, while MTIA v3, which includes HBM integration, will launch in 2026, doubling total shipment volume.
Microsoft intends to begin mass production of Maia v2 with GUC in the first half of 2026. However, the timeline for Maia v3 has been postponed due to design changes, resulting in limited ASIC shipments in the immediate future and placing Microsoft behind its competitors.
For further details on reports and market data from TrendForce’s Semiconductor Research Department, please click , or contact the Sales Department via email at .
To access more insights from TrendForce analysts regarding the most recent tech industry news, trends, and predictions, please visit .
About TrendForce
TrendForce delivers global updates, insights, and analysis for the technology industry. With over a decade of service to businesses, the company boasts a robust subscriber base of 500,000 members. TrendForce is recognized for providing perceptive and precise analysis of the tech sector through its five main research divisions: Semiconductor Research, Display Research, Optoelectronics Research, Green Energy Research, and ICT Applications Research. Established in Taipei, Taiwan, in 2000, TrendForce expanded into China in 2004, opening offices in Shenzhen and Beijing.
